How to Use ChatGPT in 2025

Hey friends,

I hope you're having a great summer start!

Many of you are using ChatGPT to plan trips and ask for recommendations to reach your dream summer body as quickly as possible. Isn’t it?

I know, because I coach many people on using AI (over 200 this year); and 90% of the time the first question I get is:

"I'm lost with too many models. What should I use for which use case? When should I use Google vs ChatGPT? If o3 is the best ChatGPT model, why use 4o at all?"

It hit me: 90% of people don't need more features — they need clarity.

So I decided to write this short brief to give you a couple of guidelines and examples showing exactly when to use reasoning models vs chat models, ChatGPT search vs Google, Deep Research, and Canvas.

Nothing advanced or fancy here. But good results usually boil down to the basic fundamentals.

How to make ChatGPT actually useful ↓↓

Choosing the Right ChatGPT Model

ChatGPT has rolled out over 1.1 updates per week this year.

That’s a lot of noise.

So here’s the simple rule:

Use a reasoning model for anything important or hard.

Use a basic chat model for speed and simplicity.

→ Examples:

• Low-stakes?

“Which fruits have the most fiber?”

→ Basic chat model is fine.

• High-stakes?

“Create a vegetarian breakfast with at least 15g fiber and 20g protein.”

→ You want a reasoning model to handle the logic, and eventually execute calculation.

• Trivia?

“Who said ‘success is never final’?”

→ Chat model gets the Winston Churchill quote right.

• Historical depth?

“Why was Churchill ousted after WWII?”

→ Use a reasoning model for nuance and layered context.

• Writing a simple email from bullet points?

→ Chat model works.

• Managing a 20-message angry thread from a client?

→ Upload it as a PDF. Let the reasoning model sort it out and draft a diplomatic reply.

I track my macros on ChatGPT - using 4o for basic entries like "Add 200g salmon" and reasoning models for complex estimates like "Add ~15 spoons rice" that need calculations.

• Chat model: 4o
• Cheap unlimited reasoning model: o4-mini
• Cheap unlimited reasoning model for coding: o4-mini-high
• Best reasoning model (limited requests): o3

Pro Tips for Reasoning Models

1. Use clear sections.

## Task
Write a summary of the following

## Key guidelines (or things to avoid)
[Add your guidelines]

## Document
[Paste content here]

2. This helps the model know what to do versus what to analyze.

3. Skip “think step by step.”

Reasoning models already think that way. Don’t confuse them.

4. Examples are optional.

They’re not always needed. Use them only when results are off.

5. Guidelines should help the model to focus on what's important especially for complex tasks;

Reasoning models can easily loop into complex reasoning on minor issues. For example when I write this prompt:

Rewrite the previous title and meta description to improve click-through rate (CTR).
Keep it under 160 characters. 
Make it persuasive, clear, and benefit-focused. 
Highlight a unique insight, use action-oriented language, and optionally include the current year. 
The audience is tech-savvy readers interested in generative AI, prompts, tools, and applications. 

If you send this the reasoning model will spend most of its "thinking capability" arguing about whether 160 characters is for both title and meta-description, or only meta-description, and generating code to calculate the number of characters, etc. While the most important task is optimizing the content.

When to Use Web Search in ChatGPT

Web search is great. But so is Google — especially when you're seeking reliability.

I used to think you must search everything using AI, until I found serious problems with AI search outputs.

Here’s how to choose:

Single fact?

Fact + explanation?

→ Examples:

• “Nvidia stock price?”

Google wins — faster and more reliable.

• “What was Nvidia’s latest earnings call? Did stock go up? Why?”

ChatGPT search shines. You’ll get the stock price, movement, and context in one reply.

• “Weather forecast in Zurich?”

Google.

• “What should I pack for Zurich, Dec 1–7?”

ChatGPT will combine the forecast with practical advice.

• “Latest global vaccination rates?”

Google works.

• “Vaccination trends over 5 years in table format?”

ChatGPT gives you the data and explains the structure.

→ Pro Tips:

• Use forward slash / to toggle search directly from the chat box.
• Use it for quick fact checks where Google feels overwhelming.

Deep Research: The Secret Weapon

This is where ChatGPT becomes your unpaid intern on performance-enhancing drugs.

ChatGPT deep research is very powerful and reliable. I use it whenever I need a well-crafted and documented analysis with many sources tackling a given topic from different angles.

Deep Research = Send it off to crawl the web, read hundreds of pages, and come back with a report.

Takes 10–20 minutes. Worth every second.

Use Cases:

• Business:
• “Compare AI chip roadmaps of Nvidia, AMD, and Intel using their latest earnings reports.”
• Boom. Detailed, structured, sourced.
• Personal Finance:
• “Find top 5 high-yield savings accounts in the US. Include hidden fees. Create a savings projection if I put $1K/month.”
• Done.
• Company Ops:
• “Compare our Q4 performance to competitors using internal files + external market data.”
• → Connect your Google Drive for even deeper analysis.

ChatGPT now supports deep research functionality on many types of sources and your own data.

When you select the Deep Research functionality you can also select the source:

Pro Tip: If you don't know how to prompt a deep research, use this custom GPT from Reddit to auto-generate deep research prompts.

Canvas: For Work That Needs Revision

Canvas is a standalone writing window that lets you edit, revise, and build on responses.

It is also very powerful for creating apps and running them live.

I personally use it when working on large documents, reports, or for generating code.

Example: Pitch document

1. I asked ChatGPT to gather all information about my company. I simply activated the "Web Search" functionality and gave it my website URL. I used o3. The results are very good.

2. Ask ChatGPT to draft a "powerful pitch document about the company for investors."

Very persuasive, but contains a lot of fluff and false information as well -__-

3. Enable Canvas, select the text you want to change and refine:

• Delete fluff.
• Add relevant information.
• Expand relevant sections with more input

4. Use inline commands:

“Rewrite this in exec-speak.”

“Add a summary.”

Canvas Tips:

• Use the back and forward buttons to jump between versions.
• Use shortcuts to edit the entire doc or just snippets.
• Export final output in Markdown and upload to Google Docs. Clean formatting.

TL;DR — The Playbook

• Default to reasoning models unless you need speed.
• Choose models based on complexity, not task type.
• Web search is great for fact + context.
• Deep research saves hours. Use it.
• Canvas is perfect for long-form and iterations.
• Use prompts that tell ChatGPT what to do, not how to think.

I hope this answered many of your questions on using ChatGPT features.

Next Week (most likely): “Using AI wrong damages your brain.”

Until then, use the tools. Don’t drown in them.

Wishing you a great sunny weekend.

– Charafeddine